[Bug bootstrap/53728] New: [4.6 regression] Bootstrap comparison failure (gcc/varasm.o differs) with CFLAGS="-O2 -march=pentium3"

[Bug bootstrap/53728] New: [4.6 regression] Bootstrap comparison failure (gcc/varasm.o differs) with CFLAGS="-O2 -march=pentium3"

[xarthisius.kk at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53728

             Bug #: 53728
           Summary: [4.6 regression] Bootstrap comparison failure
                    (gcc/varasm.o differs) with CFLAGS="-O2
                    -march=pentium3"
    Classification: Unclassified
           Product: gcc
           Version: 4.6.3
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: bootstrap
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: xarthisius.kk@gmail.com


Following steps are needed to reproduce:

./configure --host=i686-pc-linux-gnu --build=i686-pc-linux-gnu
--disable-altivec --disable-fixed-point --with-ppl --with-cloog
--disable-ppl-version-check --with-cloog-include=/usr/include/cloog-ppl
--enable-lto --enable-nls --without-included-gettext --with-system-zlib
--enable-obsolete --disable-werror --enable-secureplt --disable-multilib
--enable-libmudflap --disable-libssp --enable-libgomp --enable-checking=release
--disable-libgcj --with-arch=i686 --enable-languages=c,c++ --enable-shared
--enable-threads=posix --enable-__cxa_atexit --enable-clocale=gnu
--enable-targets=all
make -j24 'LDFLAGS=-Wl,-O1 -Wl,--as-needed' STAGE1_CFLAGS= 'BOOT_CFLAGS= -O2
-march=pentium3 -pipe' bootstrap-lean

Results in:
...
Comparing stages 2 and 3
warning: gcc/cc1plus-checksum.o differs
warning: gcc/cc1-checksum.o differs
Bootstrap comparison failure!
gcc/varasm.o differs

Regression was introduced with:

Turn on -fomit-frame-pointer by default for 32bit Linux/x86.

2010-08-12  H.J. Lu  <hongjiu.lu@intel.com>

   Uros Bizjak  <ubizjak@gmail.com>

    * config.gcc: Handle --enable-frame-pointer.
    * configure.ac: Add --enable-frame-pointer.
    * configure: Regenerated.
    * config/i386/i386.c (USE_IX86_FRAME_POINTER): Default to 0.
    (override_options): If not configured with --enable-frame-pointer,
    enable -fomit-frame-pointer (but not for TARGET_MACHO or when
    optimizing for size), -fasynchronous-unwind-tables and
    -maccumulate-outgoing-args by default.

http://gcc.gnu.org/viewcvs?view=revision&revision=163196

and seems to be gone with:

http://gcc.gnu.org/viewcvs?view=revision&revision=171649

[Bug bootstrap/53728] [4.6 regression] Bootstrap comparison failure (gcc/varasm.o differs) with CFLAGS="-O2 -march=pentium3"

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53728

Richard Guenther <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Target|                            |i686-pc-linux-gnu
   Target Milestone|---                         |4.6.4

[Bug bootstrap/53728] [4.6 regression] Bootstrap comparison failure (gcc/varasm.o differs) with CFLAGS="-O2 -march=pentium3"

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53728

--- Comment #1 from Uros Bizjak <ubizjak at gmail dot com> 2013-01-30 23:09:05 UTC ---
Can you try to build from latest 4.6 branch SVN?

The bootstrap works for me with:

CC="gcc -m32" ~/gcc-svn/branches/gcc-4_6-branch/configure
--host=i686-pc-linux-gnu --build=i686-pc-linux-gnu --with-arch=i686
--enable-languages=c

and

make -j 8 BOOT_CFLAGS="-O2 -march=pentium3 -pipe" bootstrap-lean

for

gcc version 4.6.4 20130127 (prerelease) [gcc-4_6-branch revision 195496] (GCC)

[Bug bootstrap/53728] [4.6 regression] Bootstrap comparison failure (gcc/varasm.o differs) with CFLAGS="-O2 -march=pentium3"

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53728

Uros Bizjak <ubizjak at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |WAITING
   Last reconfirmed|                            |2013-02-10
     Ever Confirmed|0                           |1

--- Comment #2 from Uros Bizjak <ubizjak at gmail dot com> 2013-02-10 07:57:30 UTC ---
Please test the source from the latest 4.6 SVN branch.

[Bug bootstrap/53728] [4.6 regression] Bootstrap comparison failure (gcc/varasm.o differs) with CFLAGS="-O2 -march=pentium3"

[xarthisius.kk at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53728

--- Comment #3 from Kacper Kowalik <xarthisius.kk at gmail dot com> 2013-02-15 11:45:13 UTC ---
(In reply to comment #2)
> Please test the source from the latest 4.6 SVN branch.

I've got report from user claiming that:

commit 10612cb1b937163dcb5207913814a2fa5902cfa2
Author: gccadmin <gccadmin@138bc75d-0d04-0410-961f-82ee72b054a4>
Date:   Fri Feb 15 00:17:27 2013 +0000

    Daily bump.

    git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/branches/gcc-4_6-branch@196061
138bc75d-0d04-0410-961f-82ee72b054a

Still fails

[Bug bootstrap/53728] [4.6 regression] Bootstrap comparison failure (gcc/varasm.o differs) with CFLAGS="-O2 -march=pentium3"

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53728

--- Comment #4 from Uros Bizjak <ubizjak at gmail dot com> 2013-02-15 12:23:16 UTC ---
(In reply to comment #3)

> > Please test the source from the latest 4.6 SVN branch.
> 
> Still fails

Then I need exact steps how to trigger the failure on my x86_64-linux, with the
official version from FSF SVN. Following the procedure, described in Comment 1,
the compilation didn't fail for me.

Leaving in WAITING state until exact recipe is available.

[Bug bootstrap/53728] [4.6 regression] Bootstrap comparison failure (gcc/varasm.o differs) with CFLAGS="-O2 -march=pentium3"

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53728

--- Comment #5 from Uros Bizjak <ubizjak at gmail dot com> 2013-03-11 21:11:00 UTC ---
Created attachment 29646
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=29646
Preprocessed source of the file that miscompares during bootstrap

This is a compare-debug failure with -O2 -march=pentium3

~/gcc-build-46/prev-gcc/xgcc -B ~/gcc-build-46/prev-gcc -O2 -march=pentium3
-fcompare-debug varasm.i 
xgcc: error: varasm.i: -fcompare-debug failure (length)

Where gcc was configured with:

CC="gcc -m32" ~/gcc-svn/branches/gcc-4_6-branch/configure
--host=i686-pc-linux-gnu --build=i686-pc-linux-gnu --with-arch=i686
--enable-languages=c

[Bug bootstrap/53728] [4.6 regression] Bootstrap comparison failure (gcc/varasm.o differs) with CFLAGS="-O2 -march=pentium3"

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53728

Uros Bizjak <ubizjak at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|WAITING                     |NEW

--- Comment #6 from Uros Bizjak <ubizjak at gmail dot com> 2013-03-11 21:31:26 UTC ---
(In reply to comment #5)

> This is a compare-debug failure with -O2 -march=pentium3

Also fails on gcc-4_6 x86_64 with -O2 -m32 -march=pentium3

~/gcc-build-46/prev-gcc/xgcc -B ~/gcc-build-46/prev-gcc
-O2 -march=pentium3 -fcompare-debug -m32 varasm.i 
xgcc: error: varasm.i: -fcompare-debug failure (length)

xgcc (GCC) 4.6.4 20130311 (prerelease) [gcc-4_6-branch revision 196601]

[Bug bootstrap/53728] [4.6 regression] Bootstrap comparison failure (gcc/varasm.o differs) with CFLAGS="-O2 -march=pentium3"

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53728

--- Comment #7 from Uros Bizjak <ubizjak at gmail dot com> 2013-03-11 22:43:38 UTC ---
Testcase:

--cut here--
extern int global_options;
extern unsigned char *mode_size;

extern int printf (const char *, ...);

void mergeable_constant_section (int mode, unsigned long long align)
{
  unsigned int modesize = mode_size[mode] * 8;

  if (global_options && modesize <= align && (align & (align - 1)) == 0)
    {
      unsigned int flags;

      printf ("%d", (int) (align / 8));
      flags = align / 8;
      printf ("%u", (unsigned int) flags);
    }
}
--cut here--

$ ~/gcc-build-46/gcc/xgcc -B ~/gcc-build-46/gcc -O2 -m32 -march=pentium3 \
-fcompare-debug pr53728.c
xgcc: error: pr53728.c: -fcompare-debug failure (length)

[Bug rtl-optimization/53728] [4.6 regression] Bootstrap comparison failure (gcc/varasm.o differs) with CFLAGS="-O2 -march=pentium3"

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53728

Uros Bizjak <ubizjak at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |vmakarov at gcc dot gnu.org
          Component|bootstrap                   |rtl-optimization

--- Comment #8 from Uros Bizjak <ubizjak at gmail dot com> 2013-03-11 23:13:52 UTC ---
The difference indeed starts with IRA.

Non-debug compile declares following registers as ever live:

;;  regs ever live      0[ax] 1[dx] 2[cx] 3[bx] 7[sp] 17[flags]

where debug compile declares:

;;  regs ever live      0[ax] 1[dx] 2[cx] 3[bx] 4[si] 7[sp] 17[flags]

There is additional [si] register in the later compilation.

Re-categorized as generic RTL-optimization problem and added CC.

[Bug rtl-optimization/53728] [4.6 regression] Bootstrap comparison failure (gcc/varasm.o differs) with CFLAGS="-O2 -march=pentium3"

[dirtyepic at gentoo dot org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53728

Ryan Hill <dirtyepic at gentoo dot org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |dirtyepic at gentoo dot org

--- Comment #9 from Ryan Hill <dirtyepic at gentoo dot org> 2013-04-04 05:32:32 UTC ---
Any chance of looking at this before the 4.6 branch closes?  If not we can work
around it locally.

[Bug rtl-optimization/53728] [4.6 regression] Bootstrap comparison failure (gcc/varasm.o differs) with CFLAGS="-O2 -march=pentium3"

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53728

--- Comment #10 from Uros Bizjak <ubizjak at gmail dot com> 2013-04-04 06:41:39 UTC ---
(In reply to comment #9)
> Any chance of looking at this before the 4.6 branch closes?  If not we can work
> around it locally.

I have analysed the failure, and it is in fact a compare-debug failure, where
presence of debug instructions changes the compilation slightly. The result is
*not* wrong code, but a different (and possibly slightly non-optimal) code.

However, as found in the description, the problem was indeed fixed by the big
update of a register allocator functionality [1], and it is nearly impossible
to find a small and self-contained change that fixed this particular problem.

The referred change that was identified as a cause to this problem has nothing
to do with register allocator, it just exposes the problem.

I propose to close the bug as WONTFIX for 4.6 branch.

[1] http://gcc.gnu.org/viewcvs/gcc?view=revision&revision=171649

[Bug rtl-optimization/53728] [4.6 regression] Bootstrap comparison failure (gcc/varasm.o differs) with CFLAGS="-O2 -march=pentium3"

[dirtyepic at gentoo dot org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53728

--- Comment #11 from Ryan Hill <dirtyepic at gentoo dot org> 2013-04-04 06:54:22 UTC ---
Agreed then.  Thanks for looking into it.

[Bug rtl-optimization/53728] [4.6 regression] Bootstrap comparison failure (gcc/varasm.o differs) with CFLAGS="-O2 -march=pentium3"

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53728

Uros Bizjak <ubizjak at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|                            |WONTFIX

--- Comment #12 from Uros Bizjak <ubizjak at gmail dot com> 2013-04-07 18:24:31 UTC ---
Marked as WONTFIX.